Author: Albert Silver
Date: 09:49:54 12/12/99
Go up one level in this thread
On December 12, 1999 at 09:48:31, Enrique Irazoqui wrote:
>On December 12, 1999 at 08:49:08, Albert Silver wrote:
>
>>Hi all,
>>
>>As the issue of SSDF ratings, and their comparative value with USCF or FIDE
>>ratings, has been a recurring theme and a number of threads have sprouted
>>recently, I thought I'd share my opinion (self-plagiarized) as I think it is
>>relevant and might shed some light on the matter.
>>
>>SSDF ratings: inflated or not?
>>Here's what I think: the ratings are not inflated in the least bit.
>>Sounds crazy doesn't it? But it's not. People get too caught up trying to make
>>these futile comparisons between SSDF ratings and human ratings whether USCF,
>>FIDE, or whatever. The point is, and it has been repeated very often, there
>>simply is no comparison. The only comparison possible is that both are generated
>>using Elo's rating system, but that's where it ends. Elo's system is supposed to
>>calculate, according to a point system, the probability of success between
>>opponents rated in that system. The SSDF rating list does that to perfection,
>>but it is based on the members of the SSDF only. If you put Fritz 5.32 on fast
>>hardware up against the Tasc R30 or whatnot, it will pulverize the machine. The
>>difference in SSDF ratings accurately depicts that. It has NOTHING to do with
>>FIDE or USCF ratings. The rating of Fritz, Hiarcs, or others on the SSDF rating
>>list depicts their probability of success against other programs on the SSDF
>>list, and that's it. It doesn't represent their probability of success against
>>humans because humans simply aren't a part of the testing. If you want to find
>>out how a program will do against humans then test it against humans, and then
>>you will find it's rating against them. The SSDF rating has nothing whatsoever
>>to do with that. As was pointed out, I believe the SSDF ratings pool is a pool
>>that is COMPLETELY isolated from all others and as such cannot possibly be
>>compared with them.
>>
>> Albert Silver
>
>I think that we don't know much of what we are talking about in this issue
>comp-comp vs. human comp, SSDF vs. Fide.
>
>There is an anecdote of Wittgenstein that comes to mind. One day in his class at
>Cambridge he put a problem to his students. Imagine that the Earth is perfectly
>spherical and there is a string that goes all around the equator; this string
>would be 40 million meters long. Now imagine a second concentric string only 1
>meter longer than the first, of 40000001 meters. Without math calculations, only
>from the top of your heads, intuitively, what would be the distance between both
>strings at each point? His students answered that it would be 1 / 40 million, or
>a near zero figure like this. Then Wittgenstein told them that the distance is
>almost 1/6 of a meter and that their wrong answers showed the value of words and
>intuitions. Shortly after he quit Cambridge for good and went fishing.
>
>Mind you, I also think that without intuitions, whatever that is, exact,
>verifiable thinking tends to sterility, so from my let's call it feminine
>intuition (astrologically I am the intuitive cancer, double cancer in fact, soon
>triple I guess :(, what crap this astrology), and going back to this comp-comp
>vs. human-comp discussion, I sometimes wonder. To make it short, when looking at
>the Rebel-Baburin and Rebel Sherbakov games, I "know" that the fast finders
>couldn't play as well as Rebel. Following the games with Fritz 6 was
>overwhelming evidence in this direction. On the other hand, why this alleged
>positional, human-like (?) superiority wouldn't also show up in comp-comp games,
>so "knowledgeable" computers would compensate with it for their slower tactical
>speed? Because it doesn't compensate and comp-comp is decided by tactics. Is
>this "superior" understanding only the adaptation of a program to human playing,
>with the only value of making human life more miserable in chess, and we believe
>this anthropocentric approach greater? Is there really a difference between
>comp-comp and human-comp? So what's up? I really wish we would be less of a
>flower collector and more of a botanist.
>
>Enrique
The story of Wittgenstein is cute, and I admit I would have failed miserably,
but then again I'm not a mathematician and make no claims to having any
mathematical inutition except on a very basic level.
Regarding computer chess and human chess, there is a world of difference as I
see it. You are right in highlighting intuition. It isn't so much intuition as
being some mystical aspect of the brain, but that our brain works in a way that
is fundamentally different from a computer's and that we cannot possibly (for
now in any case) have the computer thinking the way we do. We need intuition,
which in this case is our knowledge bouncing around in our brains (feel free to
tell me I'm full of it) trying to see where it applies and in the case of
various elements, what is more important, to make up for our lack of
exhaustiveness. We cannot possibly compete with a computer's thoroughness, not
because our brains are slow or inefficient, but because we cannot force it to
work as fast consciously and deliberately as it would at a subconscious level,
in "automatic". Although we have learned to try to determine what it is we are
looking at and seeing, how this is applied is usually somewhat of a mystery. For
example, in position A there are a number of positional and tactical elements.
According to MY experience, understanding, and preference, I will orient my
seeking along certain lines. Some lines may go 2 moves ahead (4 plies) and some
may go 5-6 or even more than 10 in specific endgame situations or middlegame
tactics. In position B I have the very same positional and tactical elements
that I found in position A, yet here my thinking may go along very different
lines. I could possibly try to explain them to a player, but to a computer? And
how do I know I have the truth of it? Mind you, I'm only talking about pure
calculation here, not even strategic planning which helps make up for our
computational insufficiencies but can't possibly be taught to a computer (yet).
A quick illustration on just how far this can go is a comment by GM Yussupov
saying that Karpov, being primarily a prophylactic player, probably first
approaches a position not by asking himself what he can do, but what his
opponent wants to do, and how he can stop it. If you take another extreme, Tal
probably strove to develop as quickly as possible and then tried to find
someplace to sacrifice a piece.
Computer programs work along their own same lines despite more or less knowledge
from program to program, and lesser or greater selectivity. We cannot impart to
them the way we think, nor is this necessarily desirable, but because their
similarities are great and consistent, matches between them are necessarily
subject to the same consistency in results. When players play a machine we fight
them on the same battlefield but with different weapons.
I think this was also the big problem with Deep Blue. You noticed that Rebel has
a contest awarding the best improvement in the balance of evaluation parameters.
I doubt Rebel has 1/100th the knowledge that Deep Blue had with its 6000
parameters. How on earth can one expect they had the ideal balance for all that
after only a year? By no means do I mean this as a slur to Hsu and his team. Its
just too much IMO. I think that as it was, Deep Blue could probably have been
improved by at least 100-200 Elo points in time with greater eval fine-tuning.
Albert Silver
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.